What's this about?
Yesterday's story was that an internal version of OpenAI's next major model solved 10 long-standing open problems in mathematics and theoretical computer science. Today OpenAI added the detail everyone repeated: the whole batch cost roughly $2,000 worth of tokens at GPT-5.6 Sol API rates.
What exactly is an "open problem"?
A precisely stated mathematical question that nobody — in decades of trying — has proven either way. They're the scoreboard of mathematics: progress on one is citable, checkable, and can't be faked by sounding smart, which is why AI results here carry more weight than benchmark scores.
What's the community saying?
10.4k likes in hours, quote-tweets across math and AI Twitter.
- The unit-economics framing: if novel research costs ~$200 per result, the question shifts from "can AI do research?" to "what's worth researching at that price?"
- Pushback: "$2,000 of tokens" excludes the many failed runs, human problem-selection, and verification labor around the wins.
- Mathematicians continue yesterday's thread: these are real but carefully chosen problems — selection effects matter before declaring math "solved".